slimclaw

name	slimclaw
description	Token optimization plugin for OpenClaw. Use when monitoring inference costs, analyzing cache efficiency, viewing token savings, or configuring windowing/routing optimizations. Provides /slimclaw command for metrics and a dashboard for visualization.

SlimClaw - Token Optimization Plugin

Inference optimization middleware for OpenClaw — windowing, routing, caching.

Installation

npm install slimclaw

Or clone directly:

cd ~/.openclaw/plugins
git clone https://github.com/evansantos/slimclaw.git
cd slimclaw && npm install && npm run build

Package

• npm: slimclaw
• GitHub: https://github.com/evansantos/slimclaw

Commands

/slimclaw

Shows current metrics:

• Total requests processed
• Input/output tokens
• Cache reads/writes
• Cache hit rate (%)
• Estimated token savings
• Routing decisions and tier distribution
• Model downgrade percentage

Example output:

📊 SlimClaw Metrics
───────────────────
📊 Total requests: 42
📥 Input tokens: 15,230
📤 Output tokens: 8,450
💾 Cache reads: 12,500
✍️ Cache writes: 3,200
📈 Cache hit rate: 79.6%
💰 Est. savings: ~10,000 tokens

Configuration

Create ~/.openclaw/plugins/slimclaw/slimclaw.config.json:

{
  "enabled": true,
  "mode": "shadow",
  "dashboard": {
    "enabled": true,
    "port": 3333
  },
  "windowing": {
    "enabled": true,
    "maxTokens": 100000,
    "preserveRecent": 10
  },
  "cache": {
    "enabled": true,
    "minContentLength": 1000
  },
  "routing": {
    "enabled": true,
    "defaultModel": "anthropic/claude-sonnet-4"
  }
}

Config Options

Option	Default	Description
enabled	true	Enable/disable plugin
mode	"shadow"	"shadow" (observe only) or "active" (optimize)
dashboard.enabled	true	Enable web dashboard
dashboard.port	3333	Dashboard HTTP port
windowing.enabled	true	Enable context windowing
windowing.maxTokens	100000	Max tokens before windowing
cache.enabled	true	Enable cache breakpoint injection
routing.enabled	false	Enable ClawRouter model routing
routing.allowDowngrade	true	Allow routing to cheaper models
routing.minConfidence	0.4	Min confidence for routing decisions
routing.pinnedModels	[]	Models exempt from routing
routing.tiers.*	haiku/sonnet/opus	Model mapping by complexity tier
routing.reasoningBudget	10000	Thinking token budget for reasoning
routing.pricing	undefined	Custom model pricing overrides (per 1k tokens)

Dashboard

When enabled, access at http://localhost:3333

Features:

• Real-time metrics display
• Token savings chart over time
• Cache hit rate visualization
• Cost savings estimation
• Dark theme UI

API Endpoints

Endpoint	Description
GET /	Dashboard HTML
GET /api/metrics	Current metrics JSON
GET /api/stats	Aggregated statistics
GET /api/routing-stats	Routing metrics and tier distribution
GET /health	Health check

Troubleshooting

Plugin not loading

cd ~/.openclaw/plugins/slimclaw
npm run build
openclaw gateway restart

Verify with /slimclaw command.

Dashboard not accessible

1. Check config: dashboard.enabled: true
2. Restart OpenClaw: openclaw gateway restart
3. Check port is free: lsof -i :3333
4. Try: http://localhost:3333 or http://<your-ip>:3333

Metrics showing zeros

• Metrics reset on restart
• Make some chat requests to populate data
• Check OpenClaw logs for SlimClaw hook activity

Build errors

cd ~/.openclaw/plugins/slimclaw
rm -rf dist node_modules
npm install
npm run build

Development

# Run tests
npm test

# Watch mode
npm run dev

# Build
npm run build

Migration: Heuristic to Hybrid ClawRouter

SlimClaw now uses hybrid routing combining ClawRouter as primary classifier with heuristic fallback:

What Changed

Before (Pure Heuristic):

• Simple keyword matching
• Basic message length thresholds
• Limited accuracy on edge cases

After (Hybrid ClawRouter):

• ClawRouter complexity scoring (primary)
• Heuristic fallback with circuit breaker
• Code pattern recognition
• Reasoning complexity detection
• Graceful fallback to heuristics

Configuration Migration

Update your config to enable routing:

{
  "routing": {
    "enabled": true,
    "minConfidence": 0.4,
    "allowDowngrade": true
  }
}

Custom Pricing

Override built-in model pricing to keep cost estimates accurate:

{
  "routing": {
    "enabled": true,
    "pricing": {
      "anthropic/claude-sonnet-4-20250514": { "inputPer1k": 0.003, "outputPer1k": 0.015 }
    }
  }
}

Custom pricing merges with defaults — only override what changed.

Backward Compatibility

• All existing configs continue working
• routing.enabled: false preserves old behavior
• No breaking changes to API or commands

Current Limitations

• Shadow mode only: llm_input/llm_output hooks are observation-only
• Active optimization blocked on OpenClaw #20416
• See SlimClaw #1 for tracking